arch/x86/kernel/traps.c

   1 /*
   2  *  Copyright (C) 1991, 1992  Linus Torvalds
   3  *  Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs
   4  *
   5  *  Pentium III FXSR, SSE support
   6  *      Gareth Hughes <gareth@valinux.com>, May 2000
   7  */
   8
   9 /*
  10  * Handle hardware traps and faults.
  11  */
  12 #include <linux/interrupt.h>
  13 #include <linux/kallsyms.h>
  14 #include <linux/spinlock.h>
  15 #include <linux/kprobes.h>
  16 #include <linux/uaccess.h>
  17 #include <linux/utsname.h>
  18 #include <linux/kdebug.h>
  19 #include <linux/kernel.h>
  20 #include <linux/module.h>
  21 #include <linux/ptrace.h>
  22 #include <linux/string.h>
  23 #include <linux/unwind.h>
  24 #include <linux/delay.h>
  25 #include <linux/errno.h>
  26 #include <linux/kexec.h>
  27 #include <linux/sched.h>
  28 #include <linux/timer.h>
  29 #include <linux/init.h>
  30 #include <linux/bug.h>
  31 #include <linux/nmi.h>
  32 #include <linux/mm.h>
  33 #include <linux/smp.h>
  34 #include <linux/io.h>
  35
  36 #ifdef CONFIG_EISA
  37 #include <linux/ioport.h>
  38 #include <linux/eisa.h>
  39 #endif
  40
  41 #ifdef CONFIG_MCA
  42 #include <linux/mca.h>
  43 #endif
  44
  45 #if defined(CONFIG_EDAC)
  46 #include <linux/edac.h>
  47 #endif
  48
  49 #include <asm/stacktrace.h>
  50 #include <asm/processor.h>
  51 #include <asm/debugreg.h>
  52 #include <asm/atomic.h>
  53 #include <asm/system.h>
  54 #include <asm/unwind.h>
  55 #include <asm/traps.h>
  56 #include <asm/desc.h>
  57 #include <asm/i387.h>
  58
  59 #include <mach_traps.h>
  60
  61 #ifdef CONFIG_X86_64
  62 #include <asm/pgalloc.h>
  63 #include <asm/proto.h>
  64 #include <asm/pda.h>
  65 #else
  66 #include <asm/processor-flags.h>
  67 #include <asm/arch_hooks.h>
  68 #include <asm/nmi.h>
  69 #include <asm/smp.h>
  70 #include <asm/io.h>
  71 #include <asm/traps.h>
  72
  73 #include "cpu/mcheck/mce.h"
  74
  75 DECLARE_BITMAP(used_vectors, NR_VECTORS);
  76 EXPORT_SYMBOL_GPL(used_vectors);
  77
  78 asmlinkage int system_call(void);
  79
  80 /* Do we ignore FPU interrupts ? */
  81 char ignore_fpu_irq;
  82
  83 /*
  84  * The IDT has to be page-aligned to simplify the Pentium
  85  * F0 0F bug workaround.. We have a special link segment
  86  * for this.
  87  */
  88 gate_desc idt_table[256]
  89         __attribute__((__section__(".data.idt"))) = { { { { 0, 0 } } }, };
  90 #endif
  91
  92 static int ignore_nmis;
  93
  94 static inline void conditional_sti(struct pt_regs *regs)
  95 {
  96         if (regs->flags & X86_EFLAGS_IF)
  97                 local_irq_enable();
  98 }
  99
 100 static inline void preempt_conditional_sti(struct pt_regs *regs)
 101 {
 102         inc_preempt_count();
 103         if (regs->flags & X86_EFLAGS_IF)
 104                 local_irq_enable();
 105 }
 106
 107 static inline void preempt_conditional_cli(struct pt_regs *regs)
 108 {
 109         if (regs->flags & X86_EFLAGS_IF)
 110                 local_irq_disable();
 111         dec_preempt_count();
 112 }
 113
 114 #ifdef CONFIG_X86_32
 115 static inline void
 116 die_if_kernel(const char *str, struct pt_regs *regs, long err)
 117 {
 118         if (!user_mode_vm(regs))
 119                 die(str, regs, err);
 120 }
 121
 122 /*
 123  * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
 124  * invalid offset set (the LAZY one) and the faulting thread has
 125  * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS,
 126  * we set the offset field correctly and return 1.
 127  */
 128 static int lazy_iobitmap_copy(void)
 129 {
 130         struct thread_struct *thread;
 131         struct tss_struct *tss;
 132         int cpu;
 133
 134         cpu = get_cpu();
 135         tss = &per_cpu(init_tss, cpu);
 136         thread = &current->thread;
 137
 138         if (tss->x86_tss.io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY &&
 139             thread->io_bitmap_ptr) {
 140                 memcpy(tss->io_bitmap, thread->io_bitmap_ptr,
 141                        thread->io_bitmap_max);
 142                 /*
 143                  * If the previously set map was extending to higher ports
 144                  * than the current one, pad extra space with 0xff (no access).
 145                  */
 146                 if (thread->io_bitmap_max < tss->io_bitmap_max) {
 147                         memset((char *) tss->io_bitmap +
 148                                 thread->io_bitmap_max, 0xff,
 149                                 tss->io_bitmap_max - thread->io_bitmap_max);
 150                 }
 151                 tss->io_bitmap_max = thread->io_bitmap_max;
 152                 tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
 153                 tss->io_bitmap_owner = thread;
 154                 put_cpu();
 155
 156                 return 1;
 157         }
 158         put_cpu();
 159
 160         return 0;
 161 }
 162 #endif
 163
 164 static void __kprobes
 165 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
 166         long error_code, siginfo_t *info)
 167 {
 168         struct task_struct *tsk = current;
 169
 170 #ifdef CONFIG_X86_32
 171         if (regs->flags & X86_VM_MASK) {
 172                 /*
 173                  * traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
 174                  * On nmi (interrupt 2), do_trap should not be called.
 175                  */
 176                 if (trapnr < 6)
 177                         goto vm86_trap;
 178                 goto trap_signal;
 179         }
 180 #endif
 181
 182         if (!user_mode(regs))
 183                 goto kernel_trap;
 184
 185 #ifdef CONFIG_X86_32
 186 trap_signal:
 187 #endif
 188         /*
 189          * We want error_code and trap_no set for userspace faults and
 190          * kernelspace faults which result in die(), but not
 191          * kernelspace faults which are fixed up.  die() gives the
 192          * process no chance to handle the signal and notice the
 193          * kernel fault information, so that won't result in polluting
 194          * the information about previously queued, but not yet
 195          * delivered, faults.  See also do_general_protection below.
 196          */
 197         tsk->thread.error_code = error_code;
 198         tsk->thread.trap_no = trapnr;
 199
 200 #ifdef CONFIG_X86_64
 201         if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
 202             printk_ratelimit()) {
 203                 printk(KERN_INFO
 204                        "%s[%d] trap %s ip:%lx sp:%lx error:%lx",
 205                        tsk->comm, tsk->pid, str,
 206                        regs->ip, regs->sp, error_code);
 207                 print_vma_addr(" in ", regs->ip);
 208                 printk("\n");
 209         }
 210 #endif
 211
 212         if (info)
 213                 force_sig_info(signr, info, tsk);
 214         else
 215                 force_sig(signr, tsk);
 216         return;
 217
 218 kernel_trap:
 219         if (!fixup_exception(regs)) {
 220                 tsk->thread.error_code = error_code;
 221                 tsk->thread.trap_no = trapnr;
 222                 die(str, regs, error_code);
 223         }
 224         return;
 225
 226 #ifdef CONFIG_X86_32
 227 vm86_trap:
 228         if (handle_vm86_trap((struct kernel_vm86_regs *) regs,
 229                                                 error_code, trapnr))
 230                 goto trap_signal;
 231         return;
 232 #endif
 233 }
 234
 235 #define DO_ERROR(trapnr, signr, str, name)                              \
 236 dotraplinkage void do_##name(struct pt_regs *regs, long error_code)     \
 237 {                                                                       \
 238         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr)  \
 239                                                         == NOTIFY_STOP) \
 240                 return;                                                 \
 241         conditional_sti(regs);                                          \
 242         do_trap(trapnr, signr, str, regs, error_code, NULL);            \
 243 }
 244
 245 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr)         \
 246 dotraplinkage void do_##name(struct pt_regs *regs, long error_code)     \
 247 {                                                                       \
 248         siginfo_t info;                                                 \
 249         info.si_signo = signr;                                          \
 250         info.si_errno = 0;                                              \
 251         info.si_code = sicode;                                          \
 252         info.si_addr = (void __user *)siaddr;                           \
 253         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr)  \
 254                                                         == NOTIFY_STOP) \
 255                 return;                                                 \
 256         conditional_sti(regs);                                          \
 257         do_trap(trapnr, signr, str, regs, error_code, &info);           \
 258 }
 259
 260 DO_ERROR_INFO(0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->ip)
 261 DO_ERROR(4, SIGSEGV, "overflow", overflow)
 262 DO_ERROR(5, SIGSEGV, "bounds", bounds)
 263 DO_ERROR_INFO(6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->ip)
 264 DO_ERROR(9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
 265 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
 266 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
 267 #ifdef CONFIG_X86_32
 268 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
 269 #endif
 270 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
 271
 272 #ifdef CONFIG_X86_64
 273 /* Runs on IST stack */
 274 dotraplinkage void do_stack_segment(struct pt_regs *regs, long error_code)
 275 {
 276         if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
 277                         12, SIGBUS) == NOTIFY_STOP)
 278                 return;
 279         preempt_conditional_sti(regs);
 280         do_trap(12, SIGBUS, "stack segment", regs, error_code, NULL);
 281         preempt_conditional_cli(regs);
 282 }
 283
 284 dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
 285 {
 286         static const char str[] = "double fault";
 287         struct task_struct *tsk = current;
 288
 289         /* Return not checked because double check cannot be ignored */
 290         notify_die(DIE_TRAP, str, regs, error_code, 8, SIGSEGV);
 291
 292         tsk->thread.error_code = error_code;
 293         tsk->thread.trap_no = 8;
 294
 295         /* This is always a kernel trap and never fixable (and thus must
 296            never return). */
 297         for (;;)
 298                 die(str, regs, error_code);
 299 }
 300 #endif
 301
 302 dotraplinkage void __kprobes
 303 do_general_protection(struct pt_regs *regs, long error_code)
 304 {
 305         struct task_struct *tsk;
 306
 307         conditional_sti(regs);
 308
 309 #ifdef CONFIG_X86_32
 310         if (lazy_iobitmap_copy()) {
 311                 /* restart the faulting instruction */
 312                 return;
 313         }
 314
 315         if (regs->flags & X86_VM_MASK)
 316                 goto gp_in_vm86;
 317 #endif
 318
 319         tsk = current;
 320         if (!user_mode(regs))
 321                 goto gp_in_kernel;
 322
 323         tsk->thread.error_code = error_code;
 324         tsk->thread.trap_no = 13;
 325
 326         if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
 327                         printk_ratelimit()) {
 328                 printk(KERN_INFO
 329                         "%s[%d] general protection ip:%lx sp:%lx error:%lx",
 330                         tsk->comm, task_pid_nr(tsk),
 331                         regs->ip, regs->sp, error_code);
 332                 print_vma_addr(" in ", regs->ip);
 333                 printk("\n");
 334         }
 335
 336         force_sig(SIGSEGV, tsk);
 337         return;
 338
 339 #ifdef CONFIG_X86_32
 340 gp_in_vm86:
 341         local_irq_enable();
 342         handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
 343         return;
 344 #endif
 345
 346 gp_in_kernel:
 347         if (fixup_exception(regs))
 348                 return;
 349
 350         tsk->thread.error_code = error_code;
 351         tsk->thread.trap_no = 13;
 352         if (notify_die(DIE_GPF, "general protection fault", regs,
 353                                 error_code, 13, SIGSEGV) == NOTIFY_STOP)
 354                 return;
 355         die("general protection fault", regs, error_code);
 356 }
 357
 358 static notrace __kprobes void
 359 mem_parity_error(unsigned char reason, struct pt_regs *regs)
 360 {
 361         printk(KERN_EMERG
 362                 "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
 363                         reason, smp_processor_id());
 364
 365         printk(KERN_EMERG
 366                 "You have some hardware problem, likely on the PCI bus.\n");
 367
 368 #if defined(CONFIG_EDAC)
 369         if (edac_handler_set()) {
 370                 edac_atomic_assert_error();
 371                 return;
 372         }
 373 #endif
 374
 375         if (panic_on_unrecovered_nmi)
 376                 panic("NMI: Not continuing");
 377
 378         printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 379
 380         /* Clear and disable the memory parity error line. */
 381         reason = (reason & 0xf) | 4;
 382         outb(reason, 0x61);
 383 }
 384
 385 static notrace __kprobes void
 386 io_check_error(unsigned char reason, struct pt_regs *regs)
 387 {
 388         unsigned long i;
 389
 390         printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
 391         show_registers(regs);
 392
 393         /* Re-enable the IOCK line, wait for a few seconds */
 394         reason = (reason & 0xf) | 8;
 395         outb(reason, 0x61);
 396
 397         i = 2000;
 398         while (--i)
 399                 udelay(1000);
 400
 401         reason &= ~8;
 402         outb(reason, 0x61);
 403 }
 404
 405 static notrace __kprobes void
 406 unknown_nmi_error(unsigned char reason, struct pt_regs *regs)
 407 {
 408         if (notify_die(DIE_NMIUNKNOWN, "nmi", regs, reason, 2, SIGINT) ==
 409                         NOTIFY_STOP)
 410                 return;
 411 #ifdef CONFIG_MCA
 412         /*
 413          * Might actually be able to figure out what the guilty party
 414          * is:
 415          */
 416         if (MCA_bus) {
 417                 mca_handle_nmi();
 418                 return;
 419         }
 420 #endif
 421         printk(KERN_EMERG
 422                 "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
 423                         reason, smp_processor_id());
 424
 425         printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
 426         if (panic_on_unrecovered_nmi)
 427                 panic("NMI: Not continuing");
 428
 429         printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
 430 }
 431
 432 static notrace __kprobes void default_do_nmi(struct pt_regs *regs)
 433 {
 434         unsigned char reason = 0;
 435         int cpu;
 436
 437         cpu = smp_processor_id();
 438
 439         /* Only the BSP gets external NMIs from the system. */
 440         if (!cpu)
 441                 reason = get_nmi_reason();
 442
 443         if (!(reason & 0xc0)) {
 444                 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
 445                                                                 == NOTIFY_STOP)
 446                         return;
 447 #ifdef CONFIG_X86_LOCAL_APIC
 448                 /*
 449                  * Ok, so this is none of the documented NMI sources,
 450                  * so it must be the NMI watchdog.
 451                  */
 452                 if (nmi_watchdog_tick(regs, reason))
 453                         return;
 454                 if (!do_nmi_callback(regs, cpu))
 455                         unknown_nmi_error(reason, regs);
 456 #else
 457                 unknown_nmi_error(reason, regs);
 458 #endif
 459
 460                 return;
 461         }
 462         if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
 463                 return;
 464
 465         /* AK: following checks seem to be broken on modern chipsets. FIXME */
 466         if (reason & 0x80)
 467                 mem_parity_error(reason, regs);
 468         if (reason & 0x40)
 469                 io_check_error(reason, regs);
 470 #ifdef CONFIG_X86_32
 471         /*
 472          * Reassert NMI in case it became active meanwhile
 473          * as it's edge-triggered:
 474          */
 475         reassert_nmi();
 476 #endif
 477 }
 478
 479 dotraplinkage notrace __kprobes void
 480 do_nmi(struct pt_regs *regs, long error_code)
 481 {
 482         nmi_enter();
 483
 484         inc_irq_stat(__nmi_count);
 485
 486         if (!ignore_nmis)
 487                 default_do_nmi(regs);
 488
 489         nmi_exit();
 490 }
 491
 492 void stop_nmi(void)
 493 {
 494         acpi_nmi_disable();
 495         ignore_nmis++;
 496 }
 497
 498 void restart_nmi(void)
 499 {
 500         ignore_nmis--;
 501         acpi_nmi_enable();
 502 }
 503
 504 /* May run on IST stack. */
 505 dotraplinkage void __kprobes do_int3(struct pt_regs *regs, long error_code)
 506 {
 507 #ifdef CONFIG_KPROBES
 508         if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
 509                         == NOTIFY_STOP)
 510                 return;
 511 #else
 512         if (notify_die(DIE_TRAP, "int3", regs, error_code, 3, SIGTRAP)
 513                         == NOTIFY_STOP)
 514                 return;
 515 #endif
 516
 517         preempt_conditional_sti(regs);
 518         do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
 519         preempt_conditional_cli(regs);
 520 }
 521
 522 #ifdef CONFIG_X86_64
 523 /* Help handler running on IST stack to switch back to user stack
 524    for scheduling or signal handling. The actual stack switch is done in
 525    entry.S */
 526 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
 527 {
 528         struct pt_regs *regs = eregs;
 529         /* Did already sync */
 530         if (eregs == (struct pt_regs *)eregs->sp)
 531                 ;
 532         /* Exception from user space */
 533         else if (user_mode(eregs))
 534                 regs = task_pt_regs(current);
 535         /* Exception from kernel and interrupts are enabled. Move to
 536            kernel process stack. */
 537         else if (eregs->flags & X86_EFLAGS_IF)
 538                 regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
 539         if (eregs != regs)
 540                 *regs = *eregs;
 541         return regs;
 542 }
 543 #endif
 544
 545 /*
 546  * Our handling of the processor debug registers is non-trivial.
 547  * We do not clear them on entry and exit from the kernel. Therefore
 548  * it is possible to get a watchpoint trap here from inside the kernel.
 549  * However, the code in ./ptrace.c has ensured that the user can
 550  * only set watchpoints on userspace addresses. Therefore the in-kernel
 551  * watchpoint trap can only occur in code which is reading/writing
 552  * from user space. Such code must not hold kernel locks (since it
 553  * can equally take a page fault), therefore it is safe to call
 554  * force_sig_info even though that claims and releases locks.
 555  *
 556  * Code in ./signal.c ensures that the debug control register
 557  * is restored before we deliver any signal, and therefore that
 558  * user code runs with the correct debug control register even though
 559  * we clear it here.
 560  *
 561  * Being careful here means that we don't have to be as careful in a
 562  * lot of more complicated places (task switching can be a bit lazy
 563  * about restoring all the debug state, and ptrace doesn't have to
 564  * find every occurrence of the TF bit that could be saved away even
 565  * by user code)
 566  *
 567  * May run on IST stack.
 568  */
 569 dotraplinkage void __kprobes do_debug(struct pt_regs *regs, long error_code)
 570 {
 571         struct task_struct *tsk = current;
 572         unsigned long condition;
 573         int si_code;
 574
 575         get_debugreg(condition, 6);
 576
 577         /*
 578          * The processor cleared BTF, so don't mark that we need it set.
 579          */
 580         clear_tsk_thread_flag(tsk, TIF_DEBUGCTLMSR);
 581         tsk->thread.debugctlmsr = 0;
 582
 583         if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
 584                                                 SIGTRAP) == NOTIFY_STOP)
 585                 return;
 586
 587         /* It's safe to allow irq's after DR6 has been saved */
 588         preempt_conditional_sti(regs);
 589
 590         /* Mask out spurious debug traps due to lazy DR7 setting */
 591         if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
 592                 if (!tsk->thread.debugreg7)
 593                         goto clear_dr7;
 594         }
 595
 596 #ifdef CONFIG_X86_32
 597         if (regs->flags & X86_VM_MASK)
 598                 goto debug_vm86;
 599 #endif
 600
 601         /* Save debug status register where ptrace can see it */
 602         tsk->thread.debugreg6 = condition;
 603
 604         /*
 605          * Single-stepping through TF: make sure we ignore any events in
 606          * kernel space (but re-enable TF when returning to user mode).
 607          */
 608         if (condition & DR_STEP) {
 609                 if (!user_mode(regs))
 610                         goto clear_TF_reenable;
 611         }
 612
 613         si_code = get_si_code(condition);
 614         /* Ok, finally something we can handle */
 615         send_sigtrap(tsk, regs, error_code, si_code);
 616
 617         /*
 618          * Disable additional traps. They'll be re-enabled when
 619          * the signal is delivered.
 620          */
 621 clear_dr7:
 622         set_debugreg(0, 7);
 623         preempt_conditional_cli(regs);
 624         return;
 625
 626 #ifdef CONFIG_X86_32
 627 debug_vm86:
 628         handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
 629         preempt_conditional_cli(regs);
 630         return;
 631 #endif
 632
 633 clear_TF_reenable:
 634         set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
 635         regs->flags &= ~X86_EFLAGS_TF;
 636         preempt_conditional_cli(regs);
 637         return;
 638 }
 639
 640 #ifdef CONFIG_X86_64
 641 static int kernel_math_error(struct pt_regs *regs, const char *str, int trapnr)
 642 {
 643         if (fixup_exception(regs))
 644                 return 1;
 645
 646         notify_die(DIE_GPF, str, regs, 0, trapnr, SIGFPE);
 647         /* Illegal floating point operation in the kernel */
 648         current->thread.trap_no = trapnr;
 649         die(str, regs, 0);
 650         return 0;
 651 }
 652 #endif
 653
 654 /*
 655  * Note that we play around with the 'TS' bit in an attempt to get
 656  * the correct behaviour even in the presence of the asynchronous
 657  * IRQ13 behaviour
 658  */
 659 void math_error(void __user *ip)
 660 {
 661         struct task_struct *task;
 662         siginfo_t info;
 663         unsigned short cwd, swd;
 664
 665         /*
 666          * Save the info for the exception handler and clear the error.
 667          */
 668         task = current;
 669         save_init_fpu(task);
 670         task->thread.trap_no = 16;
 671         task->thread.error_code = 0;
 672         info.si_signo = SIGFPE;
 673         info.si_errno = 0;
 674         info.si_code = __SI_FAULT;
 675         info.si_addr = ip;
 676         /*
 677          * (~cwd & swd) will mask out exceptions that are not set to unmasked
 678          * status.  0x3f is the exception bits in these regs, 0x200 is the
 679          * C1 reg you need in case of a stack fault, 0x040 is the stack
 680          * fault bit.  We should only be taking one exception at a time,
 681          * so if this combination doesn't produce any single exception,
 682          * then we have a bad program that isn't synchronizing its FPU usage
 683          * and it will suffer the consequences since we won't be able to
 684          * fully reproduce the context of the exception
 685          */
 686         cwd = get_fpu_cwd(task);
 687         swd = get_fpu_swd(task);
 688         switch (swd & ~cwd & 0x3f) {
 689         case 0x000: /* No unmasked exception */
 690 #ifdef CONFIG_X86_32
 691                 return;
 692 #endif
 693         default: /* Multiple exceptions */
 694                 break;
 695         case 0x001: /* Invalid Op */
 696                 /*
 697                  * swd & 0x240 == 0x040: Stack Underflow
 698                  * swd & 0x240 == 0x240: Stack Overflow
 699                  * User must clear the SF bit (0x40) if set
 700                  */
 701                 info.si_code = FPE_FLTINV;
 702                 break;
 703         case 0x002: /* Denormalize */
 704         case 0x010: /* Underflow */
 705                 info.si_code = FPE_FLTUND;
 706                 break;
 707         case 0x004: /* Zero Divide */
 708                 info.si_code = FPE_FLTDIV;
 709                 break;
 710         case 0x008: /* Overflow */
 711                 info.si_code = FPE_FLTOVF;
 712                 break;
 713         case 0x020: /* Precision */
 714                 info.si_code = FPE_FLTRES;
 715                 break;
 716         }
 717         force_sig_info(SIGFPE, &info, task);
 718 }
 719
 720 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
 721 {
 722         conditional_sti(regs);
 723
 724 #ifdef CONFIG_X86_32
 725         ignore_fpu_irq = 1;
 726 #else
 727         if (!user_mode(regs) &&
 728             kernel_math_error(regs, "kernel x87 math error", 16))
 729                 return;
 730 #endif
 731
 732         math_error((void __user *)regs->ip);
 733 }
 734
 735 static void simd_math_error(void __user *ip)
 736 {
 737         struct task_struct *task;
 738         siginfo_t info;
 739         unsigned short mxcsr;
 740
 741         /*
 742          * Save the info for the exception handler and clear the error.
 743          */
 744         task = current;
 745         save_init_fpu(task);
 746         task->thread.trap_no = 19;
 747         task->thread.error_code = 0;
 748         info.si_signo = SIGFPE;
 749         info.si_errno = 0;
 750         info.si_code = __SI_FAULT;
 751         info.si_addr = ip;
 752         /*
 753          * The SIMD FPU exceptions are handled a little differently, as there
 754          * is only a single status/control register.  Thus, to determine which
 755          * unmasked exception was caught we must mask the exception mask bits
 756          * at 0x1f80, and then use these to mask the exception bits at 0x3f.
 757          */
 758         mxcsr = get_fpu_mxcsr(task);
 759         switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
 760         case 0x000:
 761         default:
 762                 break;
 763         case 0x001: /* Invalid Op */
 764                 info.si_code = FPE_FLTINV;
 765                 break;
 766         case 0x002: /* Denormalize */
 767         case 0x010: /* Underflow */
 768                 info.si_code = FPE_FLTUND;
 769                 break;
 770         case 0x004: /* Zero Divide */
 771                 info.si_code = FPE_FLTDIV;
 772                 break;
 773         case 0x008: /* Overflow */
 774                 info.si_code = FPE_FLTOVF;
 775                 break;
 776         case 0x020: /* Precision */
 777                 info.si_code = FPE_FLTRES;
 778                 break;
 779         }
 780         force_sig_info(SIGFPE, &info, task);
 781 }
 782
 783 dotraplinkage void
 784 do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
 785 {
 786         conditional_sti(regs);
 787
 788 #ifdef CONFIG_X86_32
 789         if (cpu_has_xmm) {
 790                 /* Handle SIMD FPU exceptions on PIII+ processors. */
 791                 ignore_fpu_irq = 1;
 792                 simd_math_error((void __user *)regs->ip);
 793                 return;
 794         }
 795         /*
 796          * Handle strange cache flush from user space exception
 797          * in all other cases.  This is undocumented behaviour.
 798          */
 799         if (regs->flags & X86_VM_MASK) {
 800                 handle_vm86_fault((struct kernel_vm86_regs *)regs, error_code);
 801                 return;
 802         }
 803         current->thread.trap_no = 19;
 804         current->thread.error_code = error_code;
 805         die_if_kernel("cache flush denied", regs, error_code);
 806         force_sig(SIGSEGV, current);
 807 #else
 808         if (!user_mode(regs) &&
 809                         kernel_math_error(regs, "kernel simd math error", 19))
 810                 return;
 811         simd_math_error((void __user *)regs->ip);
 812 #endif
 813 }
 814
 815 dotraplinkage void
 816 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
 817 {
 818         conditional_sti(regs);
 819 #if 0
 820         /* No need to warn about this any longer. */
 821         printk(KERN_INFO "Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
 822 #endif
 823 }
 824
 825 #ifdef CONFIG_X86_32
 826 unsigned long patch_espfix_desc(unsigned long uesp, unsigned long kesp)
 827 {
 828         struct desc_struct *gdt = get_cpu_gdt_table(smp_processor_id());
 829         unsigned long base = (kesp - uesp) & -THREAD_SIZE;
 830         unsigned long new_kesp = kesp - base;
 831         unsigned long lim_pages = (new_kesp | (THREAD_SIZE - 1)) >> PAGE_SHIFT;
 832         __u64 desc = *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS];
 833
 834         /* Set up base for espfix segment */
 835         desc &= 0x00f0ff0000000000ULL;
 836         desc |= ((((__u64)base) << 16) & 0x000000ffffff0000ULL) |
 837                 ((((__u64)base) << 32) & 0xff00000000000000ULL) |
 838                 ((((__u64)lim_pages) << 32) & 0x000f000000000000ULL) |
 839                 (lim_pages & 0xffff);
 840         *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS] = desc;
 841
 842         return new_kesp;
 843 }
 844 #else
 845 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
 846 {
 847 }
 848
 849 asmlinkage void __attribute__((weak)) mce_threshold_interrupt(void)
 850 {
 851 }
 852 #endif
 853
 854 /*
 855  * 'math_state_restore()' saves the current math information in the
 856  * old math state array, and gets the new ones from the current task
 857  *
 858  * Careful.. There are problems with IBM-designed IRQ13 behaviour.
 859  * Don't touch unless you *really* know how it works.
 860  *
 861  * Must be called with kernel preemption disabled (in this case,
 862  * local interrupts are disabled at the call-site in entry.S).
 863  */
 864 asmlinkage void math_state_restore(void)
 865 {
 866         struct thread_info *thread = current_thread_info();
 867         struct task_struct *tsk = thread->task;
 868
 869         if (!tsk_used_math(tsk)) {
 870                 local_irq_enable();
 871                 /*
 872                  * does a slab alloc which can sleep
 873                  */
 874                 if (init_fpu(tsk)) {
 875                         /*
 876                          * ran out of memory!
 877                          */
 878                         do_group_exit(SIGKILL);
 879                         return;
 880                 }
 881                 local_irq_disable();
 882         }
 883
 884         clts();                         /* Allow maths ops (or we recurse) */
 885 #ifdef CONFIG_X86_32
 886         restore_fpu(tsk);
 887 #else
 888         /*
 889          * Paranoid restore. send a SIGSEGV if we fail to restore the state.
 890          */
 891         if (unlikely(restore_fpu_checking(tsk))) {
 892                 stts();
 893                 force_sig(SIGSEGV, tsk);
 894                 return;
 895         }
 896 #endif
 897         thread->status |= TS_USEDFPU;   /* So we fnsave on switch_to() */
 898         tsk->fpu_counter++;
 899 }
 900 EXPORT_SYMBOL_GPL(math_state_restore);
 901
 902 #ifndef CONFIG_MATH_EMULATION
 903 asmlinkage void math_emulate(long arg)
 904 {
 905         printk(KERN_EMERG
 906                 "math-emulation not enabled and no coprocessor found.\n");
 907         printk(KERN_EMERG "killing %s.\n", current->comm);
 908         force_sig(SIGFPE, current);
 909         schedule();
 910 }
 911 #endif /* CONFIG_MATH_EMULATION */
 912
 913 dotraplinkage void __kprobes
 914 do_device_not_available(struct pt_regs *regs, long error)
 915 {
 916 #ifdef CONFIG_X86_32
 917         if (read_cr0() & X86_CR0_EM) {
 918                 conditional_sti(regs);
 919                 math_emulate(0);
 920         } else {
 921                 math_state_restore(); /* interrupts still off */
 922                 conditional_sti(regs);
 923         }
 924 #else
 925         math_state_restore();
 926 #endif
 927 }
 928
 929 #ifdef CONFIG_X86_32
 930 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
 931 {
 932         siginfo_t info;
 933         local_irq_enable();
 934
 935         info.si_signo = SIGILL;
 936         info.si_errno = 0;
 937         info.si_code = ILL_BADSTK;
 938         info.si_addr = 0;
 939         if (notify_die(DIE_TRAP, "iret exception",
 940                         regs, error_code, 32, SIGILL) == NOTIFY_STOP)
 941                 return;
 942         do_trap(32, SIGILL, "iret exception", regs, error_code, &info);
 943 }
 944 #endif
 945
 946 void __init trap_init(void)
 947 {
 948 #ifdef CONFIG_X86_32
 949         int i;
 950 #endif
 951
 952 #ifdef CONFIG_EISA
 953         void __iomem *p = early_ioremap(0x0FFFD9, 4);
 954
 955         if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
 956                 EISA_bus = 1;
 957         early_iounmap(p, 4);
 958 #endif
 959
 960         set_intr_gate(0, &divide_error);
 961         set_intr_gate_ist(1, &debug, DEBUG_STACK);
 962         set_intr_gate_ist(2, &nmi, NMI_STACK);
 963         /* int3 can be called from all */
 964         set_system_intr_gate_ist(3, &int3, DEBUG_STACK);
 965         /* int4 can be called from all */
 966         set_system_intr_gate(4, &overflow);
 967         set_intr_gate(5, &bounds);
 968         set_intr_gate(6, &invalid_op);
 969         set_intr_gate(7, &device_not_available);
 970 #ifdef CONFIG_X86_32
 971         set_task_gate(8, GDT_ENTRY_DOUBLEFAULT_TSS);
 972 #else
 973         set_intr_gate_ist(8, &double_fault, DOUBLEFAULT_STACK);
 974 #endif
 975         set_intr_gate(9, &coprocessor_segment_overrun);
 976         set_intr_gate(10, &invalid_TSS);
 977         set_intr_gate(11, &segment_not_present);
 978         set_intr_gate_ist(12, &stack_segment, STACKFAULT_STACK);
 979         set_intr_gate(13, &general_protection);
 980         set_intr_gate(14, &page_fault);
 981         set_intr_gate(15, &spurious_interrupt_bug);
 982         set_intr_gate(16, &coprocessor_error);
 983         set_intr_gate(17, &alignment_check);
 984 #ifdef CONFIG_X86_MCE
 985         set_intr_gate_ist(18, &machine_check, MCE_STACK);
 986 #endif
 987         set_intr_gate(19, &simd_coprocessor_error);
 988
 989 #ifdef CONFIG_IA32_EMULATION
 990         set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
 991 #endif
 992
 993 #ifdef CONFIG_X86_32
 994         if (cpu_has_fxsr) {
 995                 printk(KERN_INFO "Enabling fast FPU save and restore... ");
 996                 set_in_cr4(X86_CR4_OSFXSR);
 997                 printk("done.\n");
 998         }
 999         if (cpu_has_xmm) {
1000                 printk(KERN_INFO
1001                         "Enabling unmasked SIMD FPU exception support... ");
1002                 set_in_cr4(X86_CR4_OSXMMEXCPT);
1003                 printk("done.\n");
1004         }
1005
1006         set_system_trap_gate(SYSCALL_VECTOR, &system_call);
1007
1008         /* Reserve all the builtin and the syscall vector: */
1009         for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
1010                 set_bit(i, used_vectors);
1011
1012         set_bit(SYSCALL_VECTOR, used_vectors);
1013 #endif
1014         /*
1015          * Should be a barrier for any external CPU state:
1016          */
1017         cpu_init();
1018
1019 #ifdef CONFIG_X86_32
1020         trap_init_hook();
1021 #endif
1022 }